How Deepsky got acquired by Airtable just 4 months out of stealth
The inside story of the superagent Airtable grabbed before anyone else understood what it really was.
Hey everybody, welcome to PMF 🔥
Today, I want to share one of the stories that every founder out there has ever wondered how they can happen.
Deepsky was acquired by Airtable just 4 months after their launch.
TL;DR
DeepSky wasn’t acquired for flashy demos or better prompts.
It was acquired because it tackled the hardest part of work before execution, turning messy questions into clear, research-backed decisions.
Airtable already runs the execution. DeepSky owned what comes before.
My honest opinion?
This is a sick product. The biggest value is the fact that you know how accurate it is, you can see where the information comes from, how conclusions are formed, and what’s grounded in real data. In that sense, it feels closer to Perplexity on accuracy.
The outputs are super strong: clean charts, clear diagrams, structured analysis, all backed by good sources. Just a really good product that produces output that is fully ready and polished, not the typical AI ChatGPT slop content we all know.
This happened at the same time as Airtable brought in a new CTO from OpenAI.
This is how it went down:
Don’t miss this week:
1. The founders had already lived inside “serious” AI
Before DeepSky existed, its three founders were already building AI where bad decisions are expensive.
Chris Chang (CEO) spent years running machine-learning products at Pinterest and then leading Studio AI at Netflix, the systems that help decide how to deploy $10B+ in content spend and support creative and commercial decisions at scale.
Mark Kim-Huang (Chief Architect) had a decade in quant finance, then senior ML roles at Box and Splunk, shipping anomaly detection, time-series forecasting, and ML infrastructure into real enterprise environments.
Forrest Moret (CTO) came out of Google and table.ai, building large-scale ML pipelines and some of the earliest very-long-context open-weight language models, with context windows stretching past a million tokens.
All three had seen the same pattern from different angles:
Organizations are drowning in fragmented data.
The highest-leverage work isn’t typing; it’s synthesizing, deciding, and planning.
And most AI demos optimize for “wow” instead of “what actually runs the business.”
DeepSky is what happens when that pattern recognition turns into a company.
So… I’ve spoken with the founders directly, and they’re open to giving the PMF community free access codes if you request one.
If you’re curious to see how this works in practice, you can play around with it directly.
You can check it out here:
You can try it for free. And if you want to go deeper, you can request a code for 3 months of Pro (600 AI credits).
2. From infra to intelligence: the long road to a “superagent”
DeepSky didn’t start as the shiny “AI superagent for founders” you see today.
The early journey ran through a stack of increasingly ambitious bets:
They first built Preemo: a fine-tuning and deployment platform that worked heavily with financial and healthcare customers. The team built tools to fine-tune AI models at scale, customized them for specific industries, and open-sourced some of what they built. This was the tools layer: give enterprises better models and better ways to deploy them.
Then came Gradient ai: an agentic system for financial services: investment, trading, risk, and compliance teams. Instead of “ask a chatbot,” Gradient was closer to “give an analyst a mandate.” It handled multi-step workflows around research, evaluation, and monitoring in high-stakes environments.
You can see the pattern getting sharper:
Start where people already pay a lot for thinking (finance).
Build systems that can plan, retrieve, and reason, not just autocomplete.
Make the outputs look like things professionals actually use: memos, reports, models.
DeepSky wasn’t a fresh idea either. The company started as Gradient, raising roughly $40M.
What stayed constant throughout was a clear belief: AI should help people understand problems and make decisions, not just generate text.
DeepSky applies that idea more broadly. Instead of loosely defined “agents,” it focuses on one thing: turning complex, messy questions into clear, well-reasoned, execution-ready finished work.
That’s why the product feels different. It’s less about chatting with an AI and more about delegating thinking from start to finish.
3. What DeepSky actually does: deep research → reasoning → execution
Most AI tools start with an answer.
You enter a prompt, and the system immediately generates a response. Often well written and polished, but rarely grounded in a complete understanding of the problem.
DeepSky works differently.
It is built as a reasoning system, not a response engine. When you give DeepSky a task it treats that request as a multi-step decision problem, not a single prompt to complete.
But is it worth the hype?
I tried it out so you don’t have to:
To see how DeepSky actually behaves in the real world, I gave it a concrete task:
This is what happened, step by step
1. Plan what needs to be understood (before doing any work)
The first thing DeepSky does is not answer.
It pauses and explicitly lays out what it needs to understand before it can make a recommendation.
This planning step is visible in the UI. You see the system decide what to learn before it “thinks.”
2. Run parallel research across branches
Once the plan is set, DeepSky launches parallel research threads across every branch of the topic.
Instead of:
search → read → summarize → repeat,
it spins up multiple workstreams at the same time, such as market sizing, competitors, pricing models, demand drivers, risks, etc. Each branch runs independently, pulling from the open web and premium sources.
The value here: you get breadth and diversity of input without having to orchestrate anything yourself.
3. Synthesize and reason before producing anything
When the research phase completes, DeepSky still doesn’t jump straight to writing.
It enters a synthesis phase where findings are reconciled, assumptions are made explicit, and tradeoffs are evaluated. Quantitative data and qualitative context are considered together to form a coherent point of view.
Only once that reasoning is complete does it move forward.
And then, DeepSky produces finished outputs, including:
Research reports
Slides
Interactive webpages
With more output formats coming soon.
For this example, we focused on the research report.
4. The output: a real business plan, not a draft
The final result isn’t a brainstorm or a rough outline.
For this example, DeepSky produced a full business plan that covered:
an executive summary
market sizing and growth trajectory
TAM / SAM / SOM breakdown
competitive mapping and positioning
suggested product roadmap
pricing and GTM strategy
risks and failure modes
What stands out isn’t the formatting, it’s the coherence.
The story holds together across sections because everything traces back to the same underlying research and reasoning. You can read it like you would a memo or deck from a strong associate (not perfect, but complete enough to react to, refine, and use).
The simplest way to think about it:
ChatGPT is great for brainstorming, ideating and doing idea ping pong
DeepSky is built to deliver high quality, presentation-ready outputs
Also, outputs come as interactive reports with charts, graphs, structured sections… not just plain text.
If you’re curious, you can view the exact report discussed above here: LINK
5. Why Airtable cared – and why the deal happened so fast
Over the past decade, Airtable quietly became the operating layer for a large portion of the enterprise, powering workflows across 80% of the Fortune 100. Its strength has always been structure: schemas, governance, and app surfaces that let non-technical teams run real work.
Recently, Airtable made its AI ambitions explicit. With Omni, its AI-powered app builder, and assistants rolled out across the product, the direction was clear: Airtable wanted to be where AI and workflows meet.
But there was a gap.
Most work doesn’t begin with “I want to build an app.”
It starts with “I don’t fully understand this problem yet.”
That’s where DeepSky fit.
DeepSky meets users at the point of uncertainty, doing deep, cross-source research and structured reasoning before anyone touches a workflow. By the time an output exists, it’s already shaped into something actionable.
From Airtable’s perspective, DeepSky was the missing insight layer.
DeepSky helps users understand and reason through complex problems.
Airtable + Omni turn those decisions into workflows and apps.
Together, they form a complete loop: from question to execution.
That fit also explains the speed of the acquisition.
By the time Airtable moved, DeepSky’s superagent was already seeing organic adoption, including going viral with Stanford MBAs this summer.
And Airtable didn’t absorb it as a feature.
All three founders and 12 team members joined as a standalone unit. DeepSky became Airtable’s second core AI product alongside Omni.
An acquisition just four months out of stealth only makes sense if the product is doing something hard to replicate, something Airtable couldn’t quickly build in-house without slowing down its roadmap.
6. What this means for founders and investors
The key insight is where DeepSky chose to compete.
Most AI products chase better answers: faster responses, cleaner outputs, stronger generation. That layer is crowded.
DeepSky went one level earlier, the judgment layer. Focusing on research, synthesis, and decision-making before execution.
That’s what made it complementary to platforms like Airtable, not competitive.
It positioned itself above execution tools and alongside decision-makers. That made it complementary to platforms like Airtable rather than competitive.
This has two important implications:
For founders, it suggests there’s more leverage in building systems that clarify decisions than in building systems that accelerate execution.
For investors, it reframes what to look for. The most defensible AI companies won’t necessarily be the fastest or flashiest. They’ll be the ones that reliably turn messy inputs into decisions people trust.
7. Where this goes next
Post-acquisition, nothing fundamental changes for existing users: DeepSky remains a self-serve product, the same team is building it, and the roadmap is pointed at deeper integrations with Airtable’s bases, apps, and automations.
The practical benefit is straightforward: outputs that don’t stop at a report. Agents can read and write structured data, trigger workflows, and connect decisions directly to operational systems.
But the more interesting story is directional:
AI products that start with thinking and end in execution.
Workflows where agents and humans share a single source of truth.
Companies that are AI-native not because they “use AI,” but because their operating system is built around it.
DeepSky’s progression, from infrastructure, to focused agents, to a general reasoning layer, and now inside Airtable, shows one practical way to build toward that.
Not by chasing trends, but by solving the hardest parts of real work and making them dependable.
My honest opinion?
This is a sick product. The biggest value is the fact that you know how accurate it is, you can see where the information comes from, how conclusions are formed, and what’s grounded in real data. In that sense, it feels closer to Perplexity on accuracy.
The outputs are super strong: clean charts, clear diagrams, structured analysis, all backed by good sources. Just a really good product that produces output that is fully ready and polished, not the typical AI ChatGPT slop content we all know.
1. Why did Airtable acquire DeepSky only four months after launch?
Airtable acquired DeepSky because it solved the hardest part of work before execution: turning ambiguous, messy problems into structured, research-backed decisions. DeepSky operated at the “judgment layer,” not at the output or execution layer—making it a perfect complement to Airtable’s workflow and app-building platform.
2. What does DeepSky actually do?
DeepSky is a reasoning and research engine that transforms complex questions into well-structured, actionable outputs—such as full business plans, research reports, and strategy documents. Instead of generating quick answers, it performs planning, parallel research, synthesis, and reasoning before producing a final deliverable.
3. How is DeepSky different from ChatGPT and other AI tools?
Most AI tools generate responses immediately based on a single prompt. DeepSky instead behaves like a multi-step analyst: it plans what to investigate, runs deep research across multiple branches, synthesizes sources, evaluates tradeoffs, and then produces a polished, coherent output. It’s built for decision-making, not brainstorming.
4. What is the “judgment layer” in AI?
The judgment layer refers to the work that happens before execution: defining the problem, gathering evidence, synthesizing insights, and forming decisions. DeepSky focused on this layer—an area with high value and high defensibility—rather than competing in the saturated space of output generation.
5. Who founded DeepSky and what is their background?
DeepSky’s founders—Chris Chang, Mark Kim-Huang, and Forrest Moret—come from deep AI and enterprise backgrounds including Pinterest, Netflix, quant finance, Google, Splunk, and table.ai. All three spent years in environments where decisions are high-stake and reasoning systems must be robust.
6. What was DeepSky before it became DeepSky?
Before launching publicly, the company built Preemo (fine-tuning + deployment infra) and Gradient (agentic systems for finance). The core thesis stayed constant: build AI that supports complex reasoning and decision-making, not just text generation.
7. Why was DeepSky valuable to Airtable specifically?
Airtable excels at structure, governance, and workflow execution—what happens after decisions are made. DeepSky excels at understanding problems and structuring decisions—what happens before execution. Together, they create a full loop from “messy question → structured decision → workflow.”
8. Did the DeepSky team join Airtable or was it just a technology acquisition?
It was an acqui-team. All three founders and 12 team members joined Airtable as a standalone unit. DeepSky is now Airtable’s second core AI product alongside Omni.
9. How does DeepSky run research?
DeepSky starts by planning what must be understood. Then it launches parallel research threads across market sizing, competitors, pricing, risks, etc. It synthesizes findings before generating any output. The result is a coherent, memo-like deliverable that mirrors the workflow of an actual analyst.
10. What kind of outputs can DeepSky create?
DeepSky produces research reports, business plans, slide decks, and interactive webpages containing charts, graphs, and structured sections. These are presentation-ready deliverables, not drafts.
11. Why is DeepSky’s approach hard to replicate?
Because it required building:
planning-aware reasoning systems
multi-branch research orchestration
synthesis pipelines
high-quality, structured outputs
This is non-trivial infrastructure that most AI products skip in favor of prettier UX or faster generation.
12. What does the DeepSky acquisition say about the future of AI for enterprises?
It indicates that high-value AI will shift from “generating answers quickly” to supporting decisions reliably. Enterprises will pay most for AI that reduces uncertainty, not AI that writes text faster.
13. Will DeepSky remain a standalone product after the acquisition?
Yes. Nothing changes for existing users. DeepSky remains self-serve, and the roadmap includes deeper integrations with Airtable bases, automations, and structured data.
14. How will DeepSky integrate with Airtable’s Omni AI platform?
DeepSky will supply the reasoning and research layer, while Omni + Airtable will handle execution—turning recommendations into workflows, apps, or automations. Agents will be able to read/write structured data and operationalize decisions instantly.
15. What is the biggest lesson for founders from the DeepSky story?
The leverage is shifting toward building systems that clarify decisions—not just accelerate execution. The most defensible AI startups will own the judgment layer where trust, coherence, and reliability matter most.
16. What is the biggest lesson for investors?
The winners in AI won’t necessarily have the fanciest demos; they’ll be the companies that transform messy inputs into high-confidence decisions that enterprises can rely on. DeepSky is an example of this emerging category.